This narrative review examines current uses of artificial intelligence in paediatric cardiology and the paediatric cardiac intensive care unit (ICU). The focus is on domains where complex datasets are generated: cardiac imaging, electrocardiography (ECG), continuous bedside monitoring, predictive analytics, waveform intelligence, and postoperative outcome prediction for congenital heart disease (CHD).
The review frames AI as a set of methods—including machine learning and deep learning—aimed at improving diagnostic consistency, risk stratification, and early recognition of clinical deterioration in children with cardiac disease. It emphasizes that AI tools are being applied across diagnostic and monitoring workflows but that evidence strength varies by domain.
The authors conducted a narrative review of PubMed, MEDLINE, and Scopus for studies published from 2018 through 2026. Included studies evaluated artificial intelligence, machine learning, or deep learning in contexts relevant to paediatric cardiology, CHD, congenital heart surgery, and cardiac critical care.
Study selection prioritized applications focused on paediatric or CHD populations within the following areas: echocardiography, cardiac MRI, electrocardiography, cardiac ICU monitoring, waveform analytics, and postoperative complication prediction. Relevant reviews were also included. From each study the reviewers extracted the study population, AI method, clinical task, model inputs, outcomes, validation strategy, and major limitations.
According to the synthesis, AI development is most advanced in cardiac imaging and electrocardiography. In these areas, machine learning and deep learning approaches have been applied to tasks such as image interpretation and ECG analysis to improve diagnostic consistency and potentially streamline workflows.
However, the review notes that evidence specific to paediatric populations and to complex CHD remains less mature than adult-focused work. Pediatric and CHD-specific challenges—such as heterogeneous anatomy, small sample sizes, and diverse postoperative physiologies—limit generalizability of many existing models.
In the cardiac ICU setting, the review describes models that integrate electronic health record (EHR) variables, high-frequency physiologic data, and established risk indices. These predictive analytics efforts target important ICU events and states, including recognition or prediction of cardiac arrest, detection of low cardiac output physiology, identification of hyperlactatemia, prediction of extubation failure, and guidance for vasoactive de-escalation.
The promise of these models lies in earlier detection of deterioration and more accurate risk stratification, which could enable timely interventions. The review, however, highlights that many models are still at the evaluation stage and lack the prospective and external validation required to demonstrate clinical impact.
The review emphasizes waveform analytics and the concept of continuous physiologic intelligence as emerging areas in paediatric cardiac critical care. High-frequency physiologic signals captured at the bedside offer rich, time-series data that can inform models of dynamic physiologic states.
Waveform-based approaches aim to extract features from continuous monitoring (for example, arterial pressure waveforms, ECG tracings, and respiratory signals) to detect subtle changes preceding clinical deterioration. While promising, these approaches face challenges including data quality, standardization across devices and sites, and the need for rigorous validation in paediatric and CHD populations.
Postoperative outcome prediction is another application area covered in the review. AI methods have been evaluated to predict complications and outcomes after congenital heart surgery, using combinations of preoperative, intraoperative, and postoperative data inputs.
Despite demonstrated associations in some studies, the review underscores that predictive performance in real-world paediatric CHD practice is limited by heterogeneity of patients and procedures, small sample sizes for rarer conditions, and inconsistent validation strategies.
A central conclusion of the review is that several critical gaps remain before AI can be routinely and autonomously used in paediatric cardiac critical care. Key deficiencies include:
These gaps mean that, although AI models can show promise in retrospective or single-center evaluations, measurable clinical benefit and workflow-safe implementation have not been consistently demonstrated.
The review concludes that artificial intelligence has potential to enhance diagnostic consistency, risk stratification, and earlier recognition of deterioration in paediatric cardiac care. Areas of greatest current maturity are imaging and ECG analysis, while ICU-focused predictive analytics and waveform intelligence are active but less developed domains.
Critically, the authors emphasize that prospective validation, external validation, calibration, and careful deployment studies demonstrating measurable clinical benefit are required before AI systems can be used autonomously in paediatric cardiac critical care.
The review identifies opportunities for future research to address these implementation gaps, particularly through multicenter collaborations, standardized datasets, and prospective trials that evaluate clinical outcomes and safety during real-world deployment.